Robert TaFounder & CEO, Clarity · Building Hyper-Personalized Agents

The moat isn't the harness

29 Aug 2026 · 2 min

Watch on YouTube

1:42

In this YouTube excerpt, Robert Ta and Jonathan discuss the emerging view that "harnesses" serve as a competitive moat for AI-driven products, contrasting it with the more common narrative that the moat itself is the harness. They outline their own approach at Clarity, which centers on a stacked harness architecture that isolates individual agents, allowing them to be incubated from an untrained state to a level where an autonomous LLM judge can certify their performance without human oversight. The conversation details the process of gathering roughly 150 trace samples via VALs (validation agents), annotating them, and partitioning the data into training (15‑20%) and dev/test sets. This pipeline enables the creation of agents that reliably generate artifacts such as stories or stand‑ups, with the LLM judge providing a closed‑loop feedback mechanism. The ultimate goal is a self‑driving organization where harnesses, validated by LLM judges, continuously produce trusted outputs, positioning the harness architecture itself—not a traditional moat—as the strategic advantage.

Chapters

Jonathan, I know you've been working hard on the Clarity Harness server-side stuff and Clarity API and evolving that. One thing I throw out there as a flavor or lens for you to respond to is Hacker News, Y Combinator, OpenAI's thought leadership. They're all talking about harnesses being the moat. I think more and more people are waking up to that. We obviously have an opinionated take with the way we think about harnesses and subjective expertise and annotation, like the data science side, meaning the philosophy of subjective experience.
Yeah, I mean, I guess we've had our own opinion about like how harnesses could really benefit for the use cases that we're talking about. And we started this off with a stacked harness architecture. And this is a way that we can really kind of isolate an individual agent in a harness and like use it to incubate the agent to go from something that has no training to something that can have an LLM judge. And I think the LLM judge is really a key concept to center on in these harness activities. And it's basically saying, have we trained our agent and our prompt well enough? That we can, without human oversight, assert that it is acting well and that it's trained to like generate our stories, it's trained to generate our standups, it's trained to generate like whatever artifacts we need to in a way that's trusted. And in order to get there, we've had this process using the VALs for some time where we try to get about 150 samples of traces, annotate those traces, and then split them into datasets for training and for dev and testing. If you can use some percentage, say like 15 to 20% of your datasets as like a training set, use another set to test the LLM judge, you get to a point where you've incubated through your curation of expertise an agent that's going to work well and can be verified by a judge. And that's where the self-driving organization sort of takes off, is that we want to create these closed loops with LLM judges that tell us this is good and this is working.